Question about AI deck setup

[wdliu356]

I'm following the ai deck tutorial but met problem with getting the figure shown. I got some questions and list them below.

1. In the Update Crazyflie and AIdeck firmware step, it says I should see ESP32: I (910) SYS: Initialized but what I saw is
   . The value in the bracket is slightly different. Does this mean the flashing work is not done successfully?

2. When I'm running the "wifi steaming example", I can only get Connecting to socket on 192.168.4.1:5000... Socket connected with no image received. It seems to get stuck at data.extend(client_socket.recv(size-len(data))). Do you have any idea why this is happening?

3. I tried customizing the ai deck firmware following your documentation and I found the code stuck at open_pi_camera_himax(&camera). I think this might also be the reason for the second question. Do you have any idea why this is happening?

[wdliu356]

Some update about my tests
Everything worked fine when I attached the AI deck to a normal crazyflie 2.1. But when I attached it to a bolt, the camera cannot be turned on. Any idea for this problem?

[wdliu356]

Hi, some more updates here.
I followed that repo and tried resistor from 1k ohm to 5k ohm and it did no help.

Other than that, I retested it with no esc's or motors connected and I found it can start successfully but also got stuck after some time.
I checked other posts and found this post bitcraze/aideck-gap8-examples#106. In my case, the code got stuck in the exactly same place data.extend(client_socket.recv(size-len(data))). Such issue also happens when using a normal crazyflie 2.1 but the connection can "survive" longer than using a bolt. I tested the voltage between the Vcom pin and the GND pin. On normal crazyflie, the voltage difference is about 3.5v and for bolt it is 4.8v (This is tested when connecting 4 esc's and motors). I don't see the reason why a higher voltage can lead to a worse performance.

I did some other interesting experiment. If I only connect one motor and esc to the board. It sometimes can start successfully and get stuck in about 30s. But if two pairs are connected, then the success rate is close to 0.

[wdliu356]

I did another test and makes me even more confused. I tried powering it with a constant DC power supply and had it work alone. I tried both 3.5v and 5v and none of of the voltage could turn on the camera successfully, stuck at data.extend(client_socket.recv(size-len(data))). I could see a current reading of 0.112A and the wifi signal was all good.

[wdliu356]

Another interesting result is that the transfer speed of JPEG_ENCODING is way quicker than RAW_ENCODING.

[wdliu356]

Some more test result. I added some log in the wifi-streamer.c file and found that in most cases, the camera_task got stuck at open_pi_camera_himax(&camera). Sometimes, if I was lucky enough to get through that line, the task would get stuck at image capture process. So probably it is still the issue of the camera initialization.

About the transfer issue I mentioned yesterday, I think it might be the format problem. It is way easier to transfer JPEG than raw data.

[gemenerik]

Regarding JPEG transfer speeds, it makes sense to me that it’s faster than RAW.

Yes, you should connect the bleeding resistor between VCOM and ground. I’m sorry to hear it didn’t help, though. It’s also interesting that the bolt seems to worsen the issue.

If you want to investigate further, consider checking the voltage on VCOM (e.g. with an oscilloscope). It should ideally drop to pretty much zero before the deck starts. This is the reason for using the bleeding resistor. It’s possible that the ESCs add so much capacitance that this discharge takes longer. Again, my hardware knowledge is limited, but I hope this helps.

Thank you for sharing all this information. If you come across anything new, please let us know. We’ll need some time to look into this thoroughly, so it may take a while.